A number arrives every Monday. It is between one and five, or it is a percentage, or it is a score somebody at a software company invented and named after a promise. It went up or it went down. Somebody in your operation is going to be told about it.
Nobody in the building can name a single Guest behind it.
That is the whole problem, and everything else on this page is a consequence of it.
What A Satisfaction Score Actually Is #
A satisfaction score is a number produced by removing the Guest from the signal.
Somebody had an experience in your operation. It had a specific shape — a Tuesday at seven-fifteen, a two-top by the window, a server three shifts into learning the menu, a kitchen running a station short, a Guest who came in already having a hard day. All of it happened to a person you could name.
The score is what is left after every one of those facts is deleted. The Guest becomes a row. The row becomes an average. The average becomes a trend line, and the trend line is what you are handed.
Which produces the inference: satisfied Guests score high; therefore a high score means satisfied Guests.
The premise is roughly true and the conclusion is unusable, because you cannot act on an average. You can only act on a Guest, a shift, a cast member, and a specific thing that happened.
Volume Is Exposure, Not Engagement #
The second instrument in this family counts activity. Reach, impressions, followers, engagement rate, review volume, mentions.
None of those measure a relationship. They measure how many people were exposed to you. An operation can double every one of those in a quarter and have fewer Guests coming back, and the dashboard will read as a win the entire time.
Engagement quality is only measurable as evidence of relationship state, which means every signal has to stay attached to four things: a Guest, a visit or interaction, a cast member, and a register. Anything aggregated past that point stopped being a read.
The Five Tests That Separate A Read From Telemetry #
Resolution. Can you name the Guest behind the signal, and the shift or interaction that produced it. A comment thread, a direct message, and an exchange at a table all pass. Reach, impressions, and a composite score do not. They measure how far you reached, not the relationship.
Initiation. Who started it. Cast-initiated contact and prompted feedback measure your effort. Guest-initiated contact — an unsolicited message, a request for a specific server by name, an unprompted return with people in tow — measures the relationship. Track the ratio, not the total.
Cost to the Guest. Rank every act by what it cost the person performing it. A star rating costs nothing. Written feedback costs attention. A referral costs social capital. Bringing someone in costs their own standing. Quality rises with cost, which is exactly why engagement counts inflate while relationships flatten.
Continuity. The same identifiable Guest appearing across time and across contexts is the only signal that separates a relational cohort from a stream of one-off transactions. Frequency of the same names beats breadth of different ones, every time.
Consequence. Engagement that changes nothing downstream was not engagement. Read it against return interval, referral velocity, recovery acceptance after a failure, and cohort tenure. If activity climbs and none of those move, you are measuring entertainment.
The Register Has To Be Declared Before The Read #
Run every signal through the eating and dining split before you interpret it.
On the eating side, thin engagement is correct. A Customer transacting cleanly through a fulfillment event is not disengaged, and demanding relational signal from that occasion will make a coherent operation look broken.
On the dining side, thin engagement is a real failure and the same number is an alarm.
Identical numbers, opposite verdicts. Which means the register has to be declared before anybody interprets anything, and no platform score in existence declares it.
Why The Vendor’s Number Cannot Do This Work #
It was built to be comparable across operations. Comparability requires that everything specific to your building be stripped out. What is left cannot see your register, your model, your standard, or your Guest.
It samples the extremes. People respond when they are delighted or angry. The Guest who had a fine, unremarkable, entirely successful visit is the least likely to say anything, and that Guest is most of your business.
It has no idea what you were attempting. An operation with a deliberately spare approach and one built on high-touch attention get scored on the same scale by people carrying whatever expectation the category put in their head.
It rewards agreeableness. Cast members learn what moves the score and optimize for it, which usually means avoiding friction. Some of the best work in a dining room creates friction on purpose.
It cannot see a recovery. A failure caught and repaired in the room produces a Guest more bonded than one who never had a problem. The score cannot distinguish that from a visit where nothing went wrong, so the most valuable thing your cast does is invisible to it.
It goes dark when the vendor does. If your read of your Guests disappears the day you cancel a subscription, you never had a read. You had access to somebody else’s.
What Replaces It #
A named-Guest ledger. Not a database of everyone. The Guests who matter — frequency, tenure, who they bring, what they order, who they ask for. Owned by you, in your building, and it does not disappear when a contract does.
Return interval by cohort. How long between visits, tracked for the same identifiable people over time. This one number does more work than every satisfaction platform in the industry combined, because it is behavior instead of opinion.
Referral velocity. New Guests arriving because a named existing Guest sent them. The most expensive act a Guest can perform and the only one that proves the relationship.
Request-by-name rate. Guests asking for a specific cast member. It reads the relationship and the cast at once.
Recovery acceptance. When something goes wrong and your cast repairs it in the room, does that Guest come back. This is the single hardest test an operation faces and it never appears on a dashboard.
The cast’s own read, collected on purpose. Your people know which Guests are regulars, which ones are drifting, and which table went badly last night. That read exists already and almost no operation has a habit for capturing it. Build the habit and you have a listening system nobody can cancel.
Your own direct question, asked in the room. Not a survey link. A specific question asked by a specific person about a specific visit, and written down. Ten of those a week outperform a thousand automated responses, because each one is attached to everything the score deleted.
What Changes Tomorrow #
Pull whatever score you currently receive and set it next to two numbers you almost certainly do not have: how many Guests came back within their normal interval this period, and how many new Guests arrived because an existing Guest sent them.
You will probably not be able to produce either number. That gap is the finding. It means the thing you have been managing is a description of strangers’ opinions, and the thing that actually determines whether the operation compounds has never been measured in your building.
Then start the smallest version of the fix. Twenty names. The Guests your cast can already identify without looking anything up. Write down when each one was last in, who they ask for, and who they bring. Update it weekly off your cast’s read.
Inside a month that list will tell you more about the state of your operation than three years of platform scores did, because every line on it is attached to a person, a visit, and a cast member.
Digging Deeper #
Every term in my framework lives in the Knowledge Base: kb.jeffreysummers.com
Terms used here: [Measurement Lock-In], [Cover Blindness], [Visibility Trap], [Operator’s Visibility Problem], [Vendor Capture], [Guest Contract], [Customer Architecture], [Two Roads], [Repair Work], [Sameness Machine], [Default Architecture], [Constraint Architecture], [Transactional Cost-Plus], [Discount Confession]
More of my work on this, all of it in the Knowledge Base:
- 4.ME.4 The Metrics That Measure The Relationship — the replacement read, in full
- 4.ME.12 Transactional Instrumentation — what the instrument is actually built to do
- The [VoC] Industry Sells You The Silence Of Your Guests — the published prosecution
- 2755 Call Your Critics — the read that replaces the score
- 2630 What’s Better Than A Loyalty Program — the same argument on the adjacent instrument
- 2745 Recovery — where a real complaint goes
